Enhancing Autonomous Systems with Deep Learning-Based Object and Aerodrome Detection

This paper presents a comprehensive approach to enhancing autonomous systems through deep learningbased object and aerodrome detection, focusing on improving situational awareness in aviation. Utilizing the AeroRunway dataset, which comprises 3,880 high-resolution synthetic images from 28 airports, collected by us specifically to address the lack of open-source aerial datasets for runway recognition, we trained state-of-the-art object detection algorithms such as YOLOv8 and YOLOv10 to accurately identify aerodromes under diverse weather conditions and altitudes. The dataset's unique combination of challenging scenarios, including poor visibility and varied camera angles, supports robust training of deep learning models. The results demonstrate significant improvements in realtime detection accuracy, with models achieving a mean average precision (mAP) of 93.9% for airport detection and 82.5% for runway detection. These findings offer promising advancements for implementing of autonomous systems in aviation, with potential applications in pilot assistance and fully autonomous landing technologies. The paper also discusses future directions for integrating real world aerial imagery with synthetic datasets to further enhance detection capabilities. Additionally, this research integrates these detection systems with autonomous landing technologies. By coupling the object detection models with control systems that execute landing maneuvers based on real-time inputs, this study makes significant strides toward achieving fully autonomous landings. Enhancing the system's ability to make rapid, informed decisions in dynamic flight environments contributes to the development of safer and more efficient autonomous aviation technologies.